|
Activity Number:
|
221
|
|
Type:
|
Topic Contributed
|
|
Date/Time:
|
Tuesday, August 5, 2008 : 8:30 AM to 10:20 AM
|
|
Sponsor:
|
Business and Economics Statistics Section
|
| Abstract - #300525 |
|
Title:
|
Semiparametric Nonlinear Vector Autoregressive Time Series Models
|
|
Author(s):
|
Yehua Li*+ and Marc G. Genton
|
|
Companies:
|
The University of Georgia and University of Geneva
|
|
Address:
|
204 Statistics Building, Athens, GA, 30602,
|
|
Keywords:
|
Autoregressive ; Climate ; Multi-index ; Nonlinear ; Penalized Spline ; Vector time seris
|
|
Abstract:
|
We develop a new class of semiparametric nonlinear vector autoregressive time series models. It has a dimension reduction flavor in that the current vector only depends on linear projections of the past vectors. The nonlinear relationship with the past lags has an additive model structure to avoid the curse of dimensionality. All the nonparametric functions in the model are univariate and estimated via P-splines. We study estimation, hypothesis testing, asymptotics, selection of the order of the autoregression and of the smoothing parameters, and nonlinear forecasting under the proposed model. We perform simulation experiments to evaluate our model in various settings. We illustrate our methodology on a climate data set and show that our model provides more accurate yearly forecasts of the El Nino phenomenon, the unusual warming of water in the Pacific Ocean.
|
